Neural Vocoding on LibriTTS clean (dev)
0.0931MAEBigVGAN
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BigVGANModel Architecture=Full2022.06 | 0.0931 | 0.7796 | 4.053 | 0.3392 | 0.1013 | 96.1 | |
| BigVGAN-baseModel Architecture=Base2022.06 | 0.1546 | 0.8569 | 3.574 | 0.418 | 0.1298 | 94.75 | |
| BigVGAN-baseAblation=w/o filter2022.06 | 0.177 | 0.8838 | 3.472 | 0.4624 | 0.1354 | 94.37 | |
| BigVGAN-baseAblation=w/o filter & snake2022.06 | 0.1899 | 0.9047 | 3.351 | 0.4852 | 0.1399 | 94.01 | |
| HiFi-GANVersion=V12022.06 | 0.2579 | 0.9773 | 3.042 | 0.6257 | 0.1545 | 93.06 | |
| UnivNet-c32Configuration=c322022.06 | 0.2772 | 0.9433 | 3.31 | 0.6942 | 0.1356 | 94.35 | |
| UnivNet-c32Pre-training dataset=train-clean-360 subset, Configuration=c32+2022.06 | 0.2803 | 0.9552 | 3.348 | 0.7017 | 0.1342 | 94.33 | |
| WaveFlow-128Model size=128 channels2022.06 | 0.2839 | 1.0706 | 3.12 | 0.9632 | 0.1339 | 94.59 | |
| WaveGlow-256Model size=256 channels2022.06 | 0.5368 | 1.3238 | 3.179 | 2.3897 | 0.1423 | 94.19 | |
| SC-WaveRNN2022.06 | 0.5517 | 2.1411 | 1.774 | 1.5854 | 0.2925 | 83 |